想入门「强化学习」，最怕的是学完一堆公式，还是不知道它怎么真的跑起来。

最近看到一本很适合入门的教材：《The Little Book of Reinforcement Learning》。

它的优点是短，但没有停在概念层面。从基础 RL 到 Monte Carlo、DQN、PPO 这些算法都有覆盖，配套 GitHub 还给了 PyTorch 实现和补充推导。

我觉得这本书很适合放在今天重新看。

因为 Agent、RLHF、test-time scaling、自我改进系统这些方向，最后都会回到一个核心问题：模型如何通过反馈改进行为。

很多人现在直接从大模型 Agent 开始学，很容易只看到 prompt、工具调用和 workflow，却缺少对策略、奖励、探索和优化的底层直觉。

这本书的价值，就在于把 RL 从抽象公式拉回到一个可读、可跑、可理解的状态。

AI Agent 时代，RL 已经变成一门必修课了。

https://github.com/alxndrTL/little-book-rl…

alxndrTL/little-book-rl

Source: https://github.com/alxndrTL/little-book-rl

The Little Book of Reinforcement Learning

This is the associated GitHub page of the Little Book of Reinforcement Learning.

This book is a short introduction to Reinforcement Learning, from the basics to applied algorithms.

In this repo, along with the book itself, you can find the supplementary material of the book. More precisely:

• under the algos/ folder, the Pytorch-based implementation of the different algorihms covered in the book, from MC to PPO.
• under the supplementary/ folder, you can find detailed explanations and rigorous proofs for the dynamic programming algorithms briefly covered in the book. This is a document I wrote in 2021.

More material is subject to be added along the way in this repo.

You can print one for youself here.

Versions of the book :

• V1 (June 2026)

The book is distributed under a non-commercial Creative Commons license.